Add hang watchdog repro tests

[exe-dev-github-integration]

Adds ignored diagnostic repro tests for three hang/no-self-recovery paths observed in current main:

• LLM stream open can remain pending without a client-side timeout.
• Python worker with no timeout waits for snippet completion instead of self-recovering.
• MCP stdio call can remain pending in stdin write before the per-request response timeout is armed.

Verification run after fast-forwarding to latest origin/main:

cargo fmt --check
cargo test -p browser-use-llm repro_model_stream_open_can_remain_pending_without_client_timeout -- --ignored --nocapture
cargo test -p browser-use-python-worker repro_worker_without_timeout_waits_for_snippet_completion -- --ignored --nocapture
cargo test -p browser-use-agent repro_stdio_call_can_remain_pending_before_response_timeout_is_armed -- --ignored --nocapture
git diff --check

---

Summary by cubic

Add ignored diagnostic repro tests for three non-self-recovering hangs: LLM stream open without a client-side timeout, Python worker snippet with no timeout, and MCP stdio write blocking before the per-request timeout is armed. Implemented in browser-use-llm, browser-use-python-worker, and browser-use-agent with a stall-after-init stdio fixture and a large request to reproduce write blocking, to validate watchdog coverage and guide fixes.

^{Written for commit c61ef11. Summary will update on new commits.}